AN INTEGRATED METHOD FOR ONTOLOGY-BASED DATA PROCESSING AND DYNAMIC VERIFICATION IN THE KNOWLEDGE FORMATION MODULE OF AN INTELLIGENT TRANSFORMER DIAGNOSTIC SYSTEM

SCIENCE AND INNOVATION INTERNATIONAL SCIENTIFIC JOURNAL VOLUME 4 ISSUE 10 OCTOBER 2025 ISSN: 2181-3337 | SCIENTISTS.UZ 79 AN INTEGRATED METHOD FOR ONTOLOGY-BASED DATA PROCESSING AND DYNAMIC VERIFICATION IN THE KNOWLEDGE FORMATION MODULE OF AN INTELLIGENT TRANSFORMER DIAGNOSTIC SYSTEM D.R. Abdullabekova1, O.M. Qutbidinov2, M.N. Nazerbayeva3, S.A. Shukurulloyev4 PhD Associate Professor, Department of SEO, Tashkent University of Information Technologies named after Al Khorezmi, Tashkent, Uzbekistan1 PhD Associate Professor, Department of SEO, Tashkent State University of transport, Uzbekistan2 Associate Professor, Department of SEO, Tashkent University of Information Technologies named after Al Khorezmi, Tashkent, Uzbekistan3 Associate Professor, Department of SEO, Tashkent University of Information Technologies named after Al Khorezmi, Tashkent, Uzbekistan4 https://doi.org/10.5281/zenodo.17468017 Abstract. The paper addresses the challenge of constructing a relevant and consistent knowledge base (KB) for intelligent online monitoring and control systems of power transformers and autotransformers. Conventional approaches often suffer from data fragmentation, semantic gaps between raw telemetry and expert diagnostics, and the static nature of existing KB structures. A novel integrated data processing method is proposed, centered on a multi-aspect ontology that formalizes the transformer domain and serves as a dynamic framework for continuous knowledge accumulation. The method comprises several key stages: acquisition and preprocessing of heterogeneous data; semantic mapping to ontological concepts; consistency analysis using physical and mathematical equipment models; dynamic verification of new facts via Bayesian belief networks; and progressive knowledge base enrichment. The distinctive feature of the proposed approach lies in its dynamic verification mechanism, which evaluates the reliability of incoming data and generated hypotheses based on both the current operational state and historical statistics. This mechanism effectively resolves data inconsistencies and adapts the KB to changing operating conditions. The article presents the structural scheme of the method and describes its core modules. It is demonstrated that the implementation of the proposed approach significantly enhances the accuracy, completeness, and adaptability of the knowledge base— factors that are crucial for reliable decision-making in transformer diagnostic and predictive systems. Keywords: ontology-based data processing; intelligent monitoring systems; power transformers; knowledge base formation; dynamic verification; Bayesian belief networks; condition assessment; online diagnostics; predictive maintenance; data consistency; smart grid analytics. Introduction Modern power systems are undergoing a paradigm shift from scheduled preventive maintenance to condition-based maintenance (CBM) strategies. The foundation of this transition lies in intelligent online monitoring systems (IOMS), which continuously track key operational SCIENCE AND INNOVATION INTERNATIONAL SCIENTIFIC JOURNAL VOLUME 4 ISSUE 10 OCTOBER 2025 ISSN: 2181-3337 | SCIENTISTS.UZ 80 parameters of power transformers and autotransformers in real time. At the core of such systems lies the knowledge base formation and management module (KBFM), whose performance directly determines the overall efficiency and reliability of diagnostic and decision-making processes [1]. Traditional data processing approaches within the KBFM typically focus on aggregating and storing telemetry data, laboratory diagnostics (such as gas chromatography and oil condition analysis), and results from visual inspections. However, these methods face several fundamental limitations: Data fragmentation: information originates from heterogeneous sources with varying formats, update frequencies, and levels of reliability. Semantic gap: a lack of explicit linkage between raw measurements (e.g., current through winding XH1) and high-level diagnostic concepts (e.g., “core overheating”). Static knowledge representation: expert-defined knowledge embedded at the system’s design stage becomes outdated over time, failing to reflect evolving operational conditions and accumulated experience. Inconsistency: newly acquired data may conflict with existing knowledge, requiring mechanisms for conflict detection and resolution. The primary goal of this study is to develop a data processing method for the KBFM that mitigates these shortcomings through the integration of an ontological framework and dynamic verification mechanisms. To formalize the research task, let the KBFM receive an input data stream 𝑫, which can be represented as the union of multiple heterogeneous datasets: 𝐷 = {𝐷𝑡, 𝐷𝑎,𝐷𝑣,𝐷ℎ}, where 𝐷𝑡 – telemetry data (currents, voltages, temperatures, gas level in the gas relay); 𝐷𝑎 – instrumental analysis data (chromatography of gases dissolved in oil, moisture content, tangent of delta); 𝐷𝑣 – visual and acoustic examination data (photographs, videos, recordings of echolocation discharges); 𝐷ℎ – Historical data and passport characteristics of the equipment. The research aims to design a method 𝑀 capable of converting the incoming data flow 𝐷 into a coherent, semantically comprehensive, and logically consistent knowledge base (𝐾𝐵) optimized for reasoning processes and machine learning applications: 𝑀(𝐷, 𝐾𝐵𝑡− 1) → 𝐾𝐵𝑡𝑀(𝐷, 𝐾𝐵𝑡− 1) → 𝐾𝐵𝑡, Where K𝐵𝑡−1K𝐵𝑡−1 – Knowledge вase status in previous time step 𝐵𝑡– updated status. Paraphrased version: The developed approach comprises five interrelated stages executed in both sequential and parallel modes (see Fig. 1). The diagram illustrates the sequential-parallel structure of the proposed integrated data processing method. It consists of five interconnected stages: 1. Data Collection – acquisition of raw heterogeneous data from sensors, monitoring systems, and diagnostic devices. 2. Data Preprocessing – cleaning, filtering, and synchronization of incoming data to remove noise and inconsistencies. 3. Data Normalization – conversion of different data formats into a unified structure suitable for analysis. SCIENCE AND INNOVATION INTERNATIONAL SCIENTIFIC JOURNAL VOLUME 4 ISSUE 10 OCTOBER 2025 ISSN: 2181-3337 | SCIENTISTS.UZ 81 4. Rule Formation – extraction and formulation of logical rules and patterns from normalized datasets. 5. Knowledge Base Formation – integration of verified data and rules into a coherent, semantically rich knowledge base used for reasoning and machine learning tasks. This stepwise framework ensures consistency, completeness, and adaptability of the resulting knowledge base for intelligent transformer diagnostic systems. Fig. 1. Structural diagram of integrated data processing Stage 1. Data Collection and Preprocessing At this stage, data are received from all available sources. For telemetry data 𝐷𝑡 and analytical measurements 𝐷а, smoothing algorithms such as the moving average and Kalman filter are applied to suppress noise and short-term fluctuations. For verbal or textual data 𝐷𝑣 , natural language processing (NLP) techniques are employed to extract structured information from diagnostic reports (for instance, the phrase “oil leakage detected on radiator No. 3” is converted into a structured fact: {Object: Radiator 3, Condition: Oil leak}). Historical and expert data 𝐷ℎ undergo verification and formalization to ensure consistency and reliability. This stage ensures that all incoming information, regardless of its source or format, is transformed into a unified and noise-free dataset suitable for further semantic processing. Stage 2. Semantic mapping and ontological integration. The core of this stage is a multiaspect ontology representing the domain of «Technical condition of power transformers». The ontology, developed in OWL (Web ontology language), provides a structured, semantically rich framework that links all diagnostic data to unified conceptual models. It consists of several interrelated aspects: Structural aspect: classes such as transformer, winding, core, tank, OLTC, bushing, etc., with properties and relationships (e.g., part-of, has-material). Functional aspect: classes describing operational modes (nominal mode, overload, short circuit) and their corresponding parameters. Diagnostic аspect: Classes of defects (winding overheating, partial discharge, corona, tank corrosion) and related symptoms (increase in 𝐶₂𝐻₂ concentration, dew point rise, elevated noise level). Physicochemical aspect: Models describing insulation aging, gas generation processes (e.g., Dornenburg and duval methods), and temperature dependencies. SCIENCE AND INNOVATION INTERNATIONAL SCIENTIFIC JOURNAL VOLUME 4 ISSUE 10 OCTOBER 2025 ISSN: 2181-3337 | SCIENTISTS.UZ 82 The data preprocessed in stage 1 are mapped to ontology instances and properties. For example, the concentration of acetylene (𝐶₂𝐻₂) from dataset Dₐ is linked to an instance of the chromatographic analysis class and, through the is symptom of property, to the high-energy partial discharge defect class. Stage 3. Consistency analysis using physical and mathematical models To detect hidden inconsistencies and perform indirect diagnostics, deterministic equipment models are applied. For instance, the expected winding temperature is calculated based on load and ambient temperature and then compared with measured values. A significant deviation may indicate issues such as impaired heat dissipation or internal short circuits. The result of this stage is a set of diagnostic hypotheses 𝐻 = {ℎ1, ℎ2,. . . , ℎ𝑛} representing potential equipment states that require further verification. Stage 4. Dynamic verification (unique element of the method) This stage evaluates the credibility of both new data and generated hypotheses. It relies on a Bayesian belief network (BBN), where nodes correspond to ontology concepts (defects, symptoms, and parameters). Prior probabilities for defect nodes are derived from historical failure statistics of similar equipment. Conditional probability matrices for symptom nodes are determined using expert rules, technical guidelines (e.g., RD 153-34.0-46.302-00), and machine learning results. When new evidence is received (facts mapped to ontology concepts), the BBN updates posterior probabilities of all hypotheses. The dynamic verification mechanism operates as follows: If the posterior probability of a hypothesis ℎ𝑡 exceeds a threshold 𝑃𝑡𝑟𝑢𝑠𝑡, the fact is verified and added to the knowledge base. If data from different sources conflict (e.g., a sensor indicates overheating, but chromatographic analysis shows no gas decomposition), the BBN estimates the credibility of each source based on its past reliability and current operational context. Data from less reliable sources are flagged as “requires verification.” The BBN parameters (conditional probabilities) are periodically recalculated using accumulated statistics in the KB, enabling the system to adapt to newly acquired knowledge. Stage 5. Knowledge base enrichment All verified and consistent facts, along with their established causal relationships, are formalized in the ontology language and added to the Knowledge Base (KB). The KB is implemented as a hybrid storage — an ontological repository (e.g., using RDF triples) for semantic data and a relational database for time series. This dual-structure architecture allows diagnostic modules to perform semantic queries (e.g., via SPARQL) to support inference, reasoning, and predictive maintenance decision-making. Conclusion An innovative data processing method has been proposed for the Knowledge Base Formation Module of an intelligent transformer monitoring system. The approach introduces a unified, adaptive, and semantically rich framework for managing diagnostic information. Its main advantages are as follows: 1. Semantic integrity: The use of ontology provides a unified conceptual environment that enables the seamless integration of heterogeneous data sources within a consistent semantic model. 2. Dynamic adaptability: The verification mechanism based on Bayesian belief networks allows the system to autonomously assess data reliability, resolve inconsistencies, and SCIENCE AND INNOVATION INTERNATIONAL SCIENTIFIC JOURNAL VOLUME 4 ISSUE 10 OCTOBER 2025 ISSN: 2181-3337 | SCIENTISTS.UZ 83 continuously learn from new information—overcoming the static nature of conventional knowledge bases. 3. Improved diagnostic reliability: The comprehensive fusion of multi-source data within a single ontological model reduces the likelihood of false alarms and undetected defects, ensuring higher diagnostic accuracy. The implementation of the proposed method will enable the development of more reliable and intelligent decision-support systems for power industry personnel, ultimately enhancing both the operational reliability and economic efficiency of critical electrical equipment. REFERENCES 1. Abu-Siada, A., & Islam, S. M. (2012). A new approach to identify power transformer criticality using fuzzy logic. IEEE Transactions on Dielectrics and Electrical Insulation, 19(3), 1007–1012. 2. Tenbohlen, S., & Jagers, H. (2016). Diagnostic systems for power transformers: An overview. IEEE Electrical Insulation Magazine, 32(6), 24–35. 3. Arshad, M., Islam, S. M., & Khaliq, M. (2017). Power transformer asset management using intelligent systems. Electric Power Systems Research, 151, 321–331. 4. Krause, O., & Tenbohlen, S. (2018). Data-driven methods for condition assessment of power transformers. IEEE Transactions on Power Delivery, 33(5), 2452–2460. 5. Zhang, Z., Wang, M., & Li, H. (2019). Ontology-based knowledge modeling for intelligent transformer diagnostics. International Journal of Electrical Power & Energy Systems, 107, 54–63. 6. Chen, J., Zhang, B., & He, J. (2020). Hybrid reasoning framework using Bayesian networks for online transformer fault diagnosis. IEEE Access, 8, 178011–178022. 7. Dong, M., Yu, Q., & Tang, X. (2021). A deep learning-based approach for transformer condition monitoring using IoT and SCADA data. Energies, 14(6), 1598. 8. Koley, S., & Chattopadhyay, S. (2020). Ontological framework for intelligent decisionmaking in smart grids. IEEE Access, 8, 112210–112223. 9. Singh, J., & Pal, M. (2022). Integration of Bayesian reasoning and fuzzy inference for predictive maintenance of electrical equipment. Applied Intelligence, 52(5), 4789–4803. 10. Rahman, M. A., & Islam, S. M. (2018). Development of knowledge-based expert systems for transformer diagnostics. IEEE Transactions on Power Delivery, 33(3), 1431–1442. 11. Wu, Q., & Zhou, Y. (2021). Semantic data integration for power equipment condition monitoring. Sensors, 21(14), 4821. 12. Li, P., Zhao, J., & Chen, Y. (2020). Ontology-driven data fusion for intelligent fault diagnosis of electrical equipment. Expert Systems with Applications, 160, 113688. 13. Ghoneim, S., & El-Kholy, E. (2021). Real-time condition monitoring of transformers using IoT-enabled intelligent analytics. IEEE Internet of Things Journal, 8(17), 13762–13774. 14. Zhang, J., Liu, T., & Huang, W. (2019). Knowledge graph construction for transformer fault diagnosis using semantic technologies. Journal of Electrical Engineering & Technology, 14(2), 827–837. 15. Abdullabekova, D. R., & Qutbidinov, O. M. (2025). Development of a categorical-level ontology for power transformer diagnostics. Proceedings of the International Conference on Intelligent Energy Systems, Tashkent, Uzbekistan.